Article Text

Download PDFPDF

Lifestyle changes at middle age and mortality: a population-based prospective cohort study
  1. Paula Berstad1,2,
  2. Edoardo Botteri1,3,
  3. Inger Kristin Larsen1,
  4. Magnus Løberg4,5,
  5. Mette Kalager4,6,
  6. Øyvind Holme4,7,
  7. Michael Bretthauer4,5,6,
  8. Geir Hoff1,2,4,5
  1. 1Department of Colorectal Cancer Screening, Cancer Registry of Norway, Oslo, Norway
  2. 2Telemark Hospital, Skien, Norway
  3. 3Division of Epidemiology and Biostatistics, European Institute of Oncology, Milan, Italy
  4. 4Department of Health Management and Health Economics, Institute of Health and Society, University of Oslo, Oslo, Norway
  5. 5Department of Transplantation Medicine, K. G. Jebsen Center for Colorectal Cancer Research, Oslo University Hospital, Oslo, Norway
  6. 6Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA
  7. 7Sørlandet Hospital, Kristiansand, Norway
  1. Correspondence to Dr Paula Berstad, Department of Colorectal Cancer Screening, Cancer Registry of Norway, P.O. Box 5313 Majorstuen, Oslo 0304, Norway; paula.berstad{at}kreftregisteret.no

Footnotes

  • Contributors GH is the principal investigator of the main NORCCAP trial. PB and GH wrote the grant application. IKL conducted data collection in 2001 and 2004. PB wrote the first draft of the manuscript and is the guarantor of the study. EB conducted the data analysis. All authors discussed data analyses and interpretation and contributed to subsequent versions of the manuscript. All authors critically revised the manuscript and approved the final version of the manuscript.

  • Funding This study was funded by the South-Eastern Norway Regional Health Authority, grant 2012094.

  • Competing interests None declared.

  • Ethics approval South-Eastern Norway Research Ethics Committee and the Norwegian Data Protection Authority.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Study protocol is available from the corresponding author. Please contact the corresponding author to discuss de-identified data requests.

View Full Text

Statistics from Altmetric.com

Introduction

A healthy lifestyle is associated with a longer life expectancy on the population level.1–3 Favourable factors include non-smoking, adequate physical activity, normal body weight, diet rich in vegetables, fruit, whole grains and low in saturated fat, salt and sugar, and low-to-moderate alcohol consumption. Individuals with an unfavourable lifestyle, as measured by lifestyle indices, may have a threefold to fourfold higher mortality rate compared with individuals with a favourable lifestyle.2–7 Improved adherence to health recommendations in adult life is associated with improved health, specifically measured as lower incidence of diabetes8 and cardiovascular diseases (CVDs),9 and also cancer.10 ,11 There is, however, limited knowledge about prevention of early death when initiating lifestyle changes and adherence to a ‘recommended lifestyle’ in adult, middle-aged life. The magnitude of effect by several lifestyle changes combined as compared with single changes is unclear.

Smoking cessation and increased physical activity are single lifestyle changes associated with reduced mortality in population-based cohorts.12–15 High body mass gained before the age of 50 predicts early mortality in the general population.16 Improved dietary habits have been shown to reduce mortality in patients with established CVD.17 The effect, however, might be different in the general population.

We established a cohort of men and women aged 50–54 years sampled from the general population who were randomised to the control group (no intervention) in the Norwegian Colorectal Cancer Prevention (NORCCAP) trial.18 All individuals were asked to fill in a mailed baseline questionnaire about their lifestyle in 2001, and a follow-up questionnaire in 2004.19 After 12 years of follow-up, we examined the association between lifestyle factors and mortality. In particular, we wanted to explore whether changes in lifestyle during the first years of follow-up led to changes in mortality.

Methods

Study population

The current cohort consists of a random sample of 7000 individuals randomised to the control arm in the NORCCAP trial, a flexible sigmoidoscopy screening trial for colorectal cancer.18 ,19 In brief, all men and women aged 50–64 years residing in Telemark county and Oslo city were identified from the National Population Registry and randomised to once-only flexible sigmoidoscopy screening (screening arm) or no screening (control arm). Screening took place in 1999–2000 for individuals aged 55–64 years, and in 2001 for individuals aged 50–54 years. Of individuals aged 50–54 years, everyone randomised to the screening arm (n=6961), and a random sample of the control arm (n=7000), was invited to a lifestyle substudy. The baseline questionnaire was handed to screening attendees at the screening centre while waiting for the examination, whereas it was mailed to the individuals in the control arm. Owing to this heterogeneity between the arms added by the findings that the screening intervention had impact on mortality and on the pattern of lifestyle changes,20 ,21 we here present results only from the control arm.

The study was approved by the Regional Research Ethics Committee and the Norwegian Data Protection Authority. All respondents provided written informed consent.

Data collection

At the time of inclusion in 2001 (baseline), all included individuals were asked to complete a mailed questionnaire about their present lifestyle (see online supplementary data S1).22 In 2004, at the same time of the year, an identical questionnaire was mailed to those individuals who had completed the questionnaire in 2001. A further follow-up questionnaire was sent in 2012, but this was not used in the main analysis of the present study. The one-page questionnaire was based on previous national surveys.23 ,24 Details of the questionnaire were reported in a previous paper.20 Briefly, lifestyle variables collected by the questionnaire were body height and weight (for calculation of body mass index, BMI, kg/m2), current smoking status, frequency of physical exercise and frequency of consumption of selected food items. Also, the number of weekly hours in paid work and a self-report of chronic disease or pain severe enough to restrict normal daily activity during the past 3 years were collected.20

Supplementary data

We calculated a lifestyle score based on adherence to international and national public health recommendations.25–29 The lifestyle score ranged from 0 to 4 points, indicating the number of lifestyle recommendations that the individual adhered to:20 normal body weight (BMI<25.0 kg/m2),25 ,28 ,29 non-smoking,26 physical activity at least 20 min per day25 ,29 and a healthy diet. A healthy diet was classified as following these three dietary habits: (1) consumption of fruit, berries and vegetables with at least five courses per day,25 ,28 ,29 (2) consumption of fatty fish equivalent to at least once per week28 and (3) consumption of red and processed meat for dinner less than four times per week.27 ,29 We further classified individuals according to a cut-off at median value of the total lifestyle score into categories ‘unfavourable’ (0–1 score points) and ‘favourable’ (2–4 score points).

The study end point was all-cause, cancer and cardiovascular death. We obtained date and cause of death until 31 December 2013 from the Norwegian Cause of Death Registry. Deaths from cancer included the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) codes C00–C97. Deaths from CVD included ICD-10 codes I00–I99.

Statistical analysis

In our survival analyses, the entry date was defined as the date of completion of the baseline questionnaire when we studied the effect of baseline characteristics on mortality, and from the completion of the 2004 follow-up questionnaire when we studied the effect of changes in lifestyle on mortality. Each participant was observed from entry date until death, emigration or 31 December 2013, whichever occurred first.

We constructed survival curves using the Kaplan-Meier method. We used univariate Cox regression models to assess the unadjusted impact of each variable on mortality, and we treated ordinal variables as numeric variables in the models to test for linear trends. To assess the effect of baseline lifestyle scores and lifestyle changes on mortality, we used multivariable Cox regression models and adjusted for sex, age, weekly working hours and chronic disease or pain during the past 3 years. We verified the assumption of proportional hazards before reporting any result. HRs and 95% CIs were reported. If any of the factors used to generate the total lifestyle score was missing, the score was coded as missing. Nevertheless, when we discriminated between ‘favourable’ (2–4 score points) and ‘unfavourable’ (0–1 score points) lifestyle, we could include some cases with partial information. Results did not change when we excluded all cases with partial information (data not shown). In the analysis of changes in lifestyle, only individuals who completed baseline and follow-up questionnaires were included. In the sensitivity analyses, we explored the associations between lifestyle changes and subsequent mortality. First, we additionally adjusted for baseline lifestyle score in the analysis of change in the dichotomised lifestyle indicator in the multivariable Cox regression models. Second, we excluded early deaths after the lifestyle change follow-up period (ie, deaths occurring during the first year after the 2004 questionnaire). Third, we excluded individuals who either in the 2001 or the 2004 questionnaire reported having had a chronic disease or pain during the past 3 years. All analyses were carried out with the SAS software (SAS Institute, Cary, North Carolina, USA) and the R software (http://cran.r-project.org/). All the reported p values were based on two-sided tests, and values <0.05 were considered statistically significant.

Results

In 2001, 6886 of the invited individuals were attainable, and 4211 (61%) completed the lifestyle questionnaire. In 2004, individuals who had responded in 2001 and who were still alive and attainable (n=4093) were approached, and 3498 responded (85%). During the follow-up from 2001 until 2013 (median follow-up time 12.3 years), 226 of the 4211 baseline questionnaire responders had died (5.4%). Of these, 110 (49%) had died from cancer and 32 (14%) from CVD. Of the 3441 successful responders to the 2004 questionnaire, 149 (4.5%) died before the end of follow-up (figure 1).

Figure 1

Study flow diagram. NORCCAP, Norwegian Colorectal Cancer Prevention trial.

Lifestyle in 2001 (baseline) and mortality

Table 1 shows participant characteristics and lifestyle factors at baseline, as well as all-cause mortality and mortality due to cancer and CVD. All-cause and cardiovascular mortality was higher in men than in women. All-cause, cancer and cardiovascular mortality was higher in current smokers compared with never and former smokers, and increased by the daily number of cigarettes smoked at baseline. The number of working hours per week, the frequency of physical exercise and consumption of fruit, berries and vegetables were inversely associated with all-cause mortality. The lifestyle score was inversely associated with all-cause (figure 2), cancer and cardiovascular mortality (table 1).

Table 1

Characteristics at baseline, and all-cause and cause-specific mortality

Figure 2

Kaplan-Meier curves for all-cause mortality during the 12 years of follow-up by baseline lifestyle score. One point was added to the lifestyle score for each of the following: (A) not currently smoking; (B) BMI<25 kg/m2; (C) physical exercise once a day or more; (D) five courses of fruit, berries and vegetables per day, three courses of meat other than poultry for dinner per week or less, and one course of fatty fish per week or more. BMI, body mass index.

Each point increment in the lifestyle score was associated with 1.3 fewer deaths per 1000 person years; HR 0.79 (95% CI 0.67 to 0.94) adjusted for age, sex, weekly working hours and chronic disease or pain during the past 3 years. The association between lifestyle score and all-cause mortality was similar in men and women (p value for heterogeneity 0.840).

Lifestyle changes from 2001 to 2004 and mortality

Table 2 shows the number of individuals categorised by changes in the lifestyle score from 2001 to 2004, and deaths during the subsequent 9 years in the change categories. Increments from one point in 2001 to two points or more in 2004 were associated with reduced mortality, whereas reductions from two points in 2001 to zero or one point in 2004 were associated with increased mortality (table 2). Measured as a continuous variable, individuals with one-point increase in the lifestyle score from 2001 to 2004 had 1.2 fewer deaths per 1000 person years than those who remained at the same level; unadjusted HR 0.77 (0.59 to 1.00); adjusted HR 0.62 (95% CI 0.45 to 0.84). The adjusted HRs for cancer-related and CVD-related mortality were 0.62 (0.41 to 0.93) and 0.52 (0.22 to 1.21), respectively.

Table 2

Number of individuals by the lifestyle score* in 2001 combined with the score value in 2004, and number of deaths from 2004 to 2013

When lifestyle was measured as a dichotomised variable ‘unfavourable’ (0–1 score points) versus ‘favourable’ (2–4 score points), individuals in the ‘unfavourable’ category in 2001 who improved to the ‘favourable’ category in 2004 had 4.8 fewer deaths per 1000 person years than those who remained in the ‘unfavourable’ category; adjusted HR 0.31 (95% CI 0.13 to 0.70; figure 3A). Further, individuals in the ‘favourable’ category in 2001 who remained in the ‘favourable’ category in 2004 had 3.1 fewer deaths per 1000 person years than those who declined to the ‘unfavourable’ category in 2004; adjusted HR 0.45 (95% CI 0.22 to 0.91; figure 3B). The association between lifestyle score change and all-cause mortality was similar in men and women (p value for heterogeneity 0.400).

Figure 3

Kaplan-Meier curves for all-cause mortality by changes in the dichotomised lifestyle score. (A) Survival in individuals with an unfavourable lifestyle in 2001; (B) survival in individuals with a favourable lifestyle in 2001; unfavourable lifestyle: <2 score points; favourable lifestyle: score ≥2 points. HR, with 95% CI, was adjusted for gender, age, working hours per week and chronic disease in the past 3 years.

When comparing the ‘unfavourable to unfavourable’ and the ‘unfavourable to favourable’ groups, we found a statistically significant difference in mean baseline lifestyle score (0.71 vs 0.92 score points, p<0.001). Mean baseline lifestyle score was also significantly different between the ‘favourable to unfavourable’ and ‘favourable to favourable’ groups (2.09 vs 2.26 score points, p<0.001). In a sensitivity analysis including the baseline lifestyle score as a covariate, HRs for all-cause mortality in the change categories of the dichotomised lifestyle indicator did not change (not shown). Exclusion of early deaths after the lifestyle change follow-up period (ie, deaths occurring during the first year after the 2004 questionnaire) did not alter the results (comparing ‘unfavourable to favourable’ with ‘unfavourable to unfavourable’ groups, adjusted HR was 0.32 (95% CI 0.14 to 0.74), and comparing ‘favourable to favourable’ with ‘favourable to unfavourable’ groups, adjusted HR was 0.42 (95% CI 0.20 to 0.88)). Furthermore, when excluding individuals who reported chronic disease or pain during the past 3 years either in 2001 or 2004 (n=948), the adjusted all-cause mortality HR comparing ‘unfavourable to favourable’ with ‘unfavourable to unfavourable’ groups was 0.18 (95% CI 0.04 to 0.74), and 0.43 (95% CI 0.18 to 1.05) comparing ‘favourable to favourable’ with ‘favourable to unfavourable’ groups. Finally, for those who did not complete the 2004 follow-up questionnaire, we first used the information from the baseline questionnaire (assuming that no changes occurred), and second we used the information from the 2012 follow-up questionnaire, when available. Neither method altered the results (data not shown).

The association between changes in single lifestyle factors from 2001 to 2004 and all-cause mortality showed a similar pattern as the lifestyle scores (see online supplementary figures S1 and S2). Smoking cessation from 2001 to 2004 was associated with a 47% reduction in all-cause mortality; adjusted HR 0.53 (95% CI 0.28 to 0.99). Smoking cessation was also associated with an increase in body weight by 2 kg, compared to continuous smokers (p<0.001).

Discussion

In this prospective study in a population-based cohort, we showed that increased adherence to recommendations for a healthy lifestyle may reduce mortality by up to 69%. To our knowledge, this is the first report to show an effect on mortality after overall lifestyle changes in middle-aged adults. The present study also showed that the number of health recommendations adhered to at baseline is inversely associated with mortality. These associations highlight the potential of lifestyle and lifestyle improvement in preventing death before the age of 68 years, and therefore suggest a significant public health message. Health behavioural changes may have a substantial impact on years left in a middle-aged general population.

Lifestyle interventions in the general population have not been shown to prevent non-communicable diseases and death.30 ,31 This observational study suggests that primary prevention may, however, be successful in motivated individuals who freely change into a more favourable lifestyle. People who decide to modify their lifestyle, as opposed to having change imposed on them in an intervention study, might be different from those that do not. Observational studies focusing on single lifestyle improvements in the general population suggest that particularly smoking cessation15 ,32 and increased physical activity12 ,14 may improve life expectancy. We observed no mortality reduction by increase in physical exercise from less than once a day to at least once a day (recommended level), or if dietary habits improved. The effect of dietary changes on future mortality in the general population is poorly documented in the literature. There is, however, ecological evidence that particularly mortality from CVD follows dietary changes in a population.33 ,34 As cardiovascular mortality was low in the present population, a power to detect an association with dietary changes was likely to be low. Individuals who maintain a stable and normal body weight over years have lower mortality compared with obese individuals.16 Weight loss in middle-aged overweight or obese individuals, however, has not been associated with mortality.16 Similarly, we did not observe any difference in all-cause mortality related to weight change. We could, however, not adjust for weight changes due to disease in this study. The above associations between changes in single lifestyle factors and mortality reported in the literature have been observed in mixed-age or elderly populations in studies where overall lifestyle scores have not been applied. The present study showed that improvement from poor lifestyle, assessed by an overall score at middle age, might prevent death before age 68 years.

The effect of lifestyle improvement was on average a 38% mortality reduction by adherence to one additional health recommendation. A meta-analysis of the lifestyle at baseline and mortality found 21–31% lower mortality for each adopted beneficial lifestyle habit.3 Our results on changes in lifestyle add to this meta-analysis, suggesting that habits adopted at young age and changes at age 50 might prevent early deaths.

The criteria for achieving each lifestyle score point in the present study were rigid. An increase of one point in the lifestyle score in the follow-up questionnaire may have required a strong effort by the participant. For instance, the success rate of smoking cessation decreases by age,35 and may be very difficult after age 50. Similarly, it is easier to maintain a physically active lifestyle at middle age if it is acquired at a young age, than to increase the level of physical exercise up to the recommended daily level at middle age.36 We therefore assume that the relatively small group that improved from the ‘unfavourable’ category to the ‘favourable’ category (n=334, 19% of those with ‘unfavourable’ baseline lifestyle) consisted of highly motivated individuals. The 69% lower mortality in this group compared with those remaining in the ‘unfavourable’ category suggests that healthy lifestyle changes are highly effective among those with the poorest lifestyle in their 50s. Smoking cessation was the only effective single improvement. We could also observe opposite effects of lifestyle change; mortality was higher in the group declining from ‘favourable’ to ‘unfavourable’ compared with those who managed to maintain ‘favourable’ habits. This effect cannot be attributed to any single factor but appears as a sum of changes in the factors other than smoking, since very few baseline non-smokers started smoking during follow-up.

Of the available population from the NORCCAP trial for this study, we chose to include only the control arm not invited to colorectal cancer screening. We evaluated that the control and the intervention (colorectal cancer screening) arms were not homogeneous and therefore should not be combined or compared. We based this on the significantly higher improvement of lifestyle in the control compared with the intervention arm,20 the effect of intervention on colorectal cancer mortality21 and the handling of the baseline lifestyle questionnaire in 2001, which was different between the arms. However, the associations between lifestyle changes and mortality in the intervention arm were partly similar but weaker as those in the control arm —for example, change from unfavourable to favourable versus remaining unfavourable was significantly protective in controls but not in screenees (HR 0.77, 95% CI 0.42 to 1.42). When including both arms in the analysis, the results were the same as in the control arm alone, but weaker (HR 0.51, 95% CI 0.31 to 0.83). The baseline differences in the lifestyle score between the dichotomised groups suggest that those who improved their lifestyle from unfavourable in fact had a better lifestyle at start compared with those who did not improve. Similarly, the baseline lifestyle score was more favourable in those who kept up with their favourable lifestyle compared with those who declined from favourable to unfavourable. We cannot exclude the possibility that these baseline differences might explain some of the differences in mortality between the change groups. However, the differences in mortality between the groups remained also after adjusting for the baseline lifestyle score, suggesting independent association between changes in lifestyle and mortality.

Strengths and limitations

The high response rate at follow-up (85% of the baseline responders responded to the 3-year questionnaire) is a strength of this study. The current study population, however, might not be representative for the general Norwegian population: 5.4% of the present study cohort died during follow-up, which is lower than in the source population (8.3% of all control participants aged 50–54 years in the NORCCAP trial died during follow-up). This difference is expected, according to differences in health, health consciousness and sociodemographic profile between respondents in questionnaire-based studies and the background population.37 ,38 In the present study, the selection of a population with longer life expectancy and a more favourable lifestyle might affect the generalisability of our findings. Moreover, we could not rule out desirability bias due to misreporting in the self-report questionnaire, which might have biased our results towards the null hypothesis of no association between unfavourable lifestyle and mortality. Since we had information on only a limited number of possible confounders, residual confounding might have played a role in the analysis. Finally, we are uncertain about the extent to which it is possible to identify genuine lifestyle change with only two time points of measurement. However, a further follow-up questionnaire in 2012 showed that the trend in lifestyle changes observed between 2001 and 2004 was generally maintained.20

Having a chronic disease might affect an individual's way of life either favourably or unfavourably, including activity level and dietary habits,39 ,40 and further affect risk of early death. Such bias cannot be completely eliminated in observational studies such as the present one. When we excluded individuals who reported either in 2001 or 2004 that they had a chronic disease or pain during the past 3 years, the results did not change. The short questionnaire might not be adequate for assessing absolute levels of lifestyle factors in individuals, but appears suitable for ranking participants relative to each other, and assessing changes at a group level. Alcohol intake was not included in the questionnaire due to limited quality of self-reported alcohol intake41 and space restrictions. It would, however, be expected that low alcohol consumption is related to adherence to other lifestyle recommendations.42

Conclusion

The number of national health recommendations adhered to at age 50–54 years was inversely associated with mortality. Further, the present study highlights the importance of improvement in lifestyle at this age to prevent early death.

What is already known on this subject

  • Following a combination of public health guidelines is associated with low mortality.

  • Smoking cessation and increase in physical activity are single changes associated with improved longevity.

  • The impact of overall lifestyle changes on mortality at middle age has not been evaluated.

What this study adds

  • Improvement of lifestyle at age 50–60 years, measured as an increased number of public health guidelines a person adheres to, was associated with reduced early mortality.

  • Lifestyle changes towards adherence to public health guidelines at middle age; non-smoking, regular physical activity, healthy body weight and healthy diet at middle age may prevent early death.

Acknowledgments

The authors thank the members of the NORCCAP steering committee, Eva Skovlund, Jörn Schneede, Tor Iversen, Morten H. Vatn, Kjell Magne Tveit, Tor Jac Eide and Jon Lekven. This work was supported by the South-Eastern Norway Regional Health Authority (grant 2012094).

References

View Abstract

Footnotes

  • Contributors GH is the principal investigator of the main NORCCAP trial. PB and GH wrote the grant application. IKL conducted data collection in 2001 and 2004. PB wrote the first draft of the manuscript and is the guarantor of the study. EB conducted the data analysis. All authors discussed data analyses and interpretation and contributed to subsequent versions of the manuscript. All authors critically revised the manuscript and approved the final version of the manuscript.

  • Funding This study was funded by the South-Eastern Norway Regional Health Authority, grant 2012094.

  • Competing interests None declared.

  • Ethics approval South-Eastern Norway Research Ethics Committee and the Norwegian Data Protection Authority.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Study protocol is available from the corresponding author. Please contact the corresponding author to discuss de-identified data requests.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.